How mooter compares
We're not the only LLM router. We are the one built for Claude Code. Every score below is counted from the cells, not asserted. mooter classifies each prompt — locally, or with Haiku when an API key is set and the local rules are unsure — and routes it to a tier from local Ollama (T0) up to Opus (T3). Where a competitor's capability is undocumented we mark it partial or n/d, never invent a win.
Eleven capabilities. Mooter scores 10/11.
The capabilities below are derived from the real pain points of running many Claude Code sessions at once. Each tool does something well — none does all eleven.
| Capability | mooter v1.53.0 | Composio AO agent os | Conductor orchestr. | Cursor Bg bg agents | Agent Teams anthropic | Codex openai | Antigravity google | Termdock multiplexer |
|---|---|---|---|---|---|---|---|---|
Spawn agents mooter local by default · others cloud-only | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ |
Local-first runs without the cloud | ✓ | ◐ | ◐ | ◐ | ✓ | ◐ | ◐ | ✓ |
Cross-session $ savings reads each local session's prompts and routing tiers — no tokens | Not measured (no token logging) | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
5-hour quota forecast predicts when you hit the wall | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
Cross-session routing learning retunes the classifier from every session's routing log when you run /update-router | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
4-layer sandbox network · fs · secrets · config | ✓ | ◐ | ◐ | ✓ | ✓ | ✓ | ⚠† | ◐ |
Intent-based UX say the goal, not the model | ✓ | ✗ | ◐ | ✓ | ✗ | ✓ | ✗ | ✗ |
State-of-art install wizard one path, no foot-guns | ✓ | ◐ | ◐ | ✓ | ◐ | ✓ | ◐ | ◐ |
Multiplexer plugins Zellij · tmux · WezTerm · Warp | ✓ | ✗ | ✗ | ✗ | ◐ | ✗ | ✗ | ✓ |
Orchestration locks across terminals Worktree Conductor — no two agents on one file | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
Workflow visibility statusline chip always-on HUD of what is running | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Score | 10 / 11 | 1 / 11 | 1 / 11 | 4 / 11 | 3 / 11 | 4 / 11 | 1 / 11 | 2 / 11 |
Scores are derived honestly from the per-row cells, not curated to make Mooter look better. Mooter wins 4 capabilities no other tool has — 5h quota forecast, cross-session routing learning, orchestration locks across terminals, and the workflow-visibility statusline chip.
And against the routers and proxies
| mooter | Claude Code default | LiteLLM proxy | Continue.dev | OpenRouter | |
|---|---|---|---|---|---|
| Architecture | Hook (local) | Direct API call | HTTP proxy | IDE plugin | API gateway |
| Local models (Ollama) | ✓ T0 native | ✗ | ✓ (configurable) | ✓ | ✗ |
| Auto-routing by complexity | ✓ T0–T3 axis | ✗ Opus on all | ⚠️ rule-based | ⚠️ manual | ⚠️ tags |
| Domain routing (packs) | ✓ 7+ Moo Packs | ✗ | ✗ | ✗ | ✗ |
| Pre-prompt overhead (measured) | ✓ classify 0.001 ms · hook 177.1 ms p50 | n/a | n/d | n/d | n/d |
| Code/prompts leave machine | ⚠️ T0 routes local; cloud Haiku if key set | ✓ all to Anthropic | ✓ through proxy | ✓ to cloud | ✓ via gateway |
| Pack-based specialization | ✓ Moo Packs | ✗ | ✗ | ⚠️ commands | ✗ |
| Adapter Forge (local LoRA) | ⚠️ Wave 26 (training) | ✗ | ✗ | ✗ | ✗ |
| Subscription-aware | ✓ tier-detect | ✗ | ⚠️ via env | ✗ | n/a |
| Live statusline HUD | ✓ 3-line | ✗ | ✗ | ⚠️ side panel | ✗ |
| Per-bash tool badge | ✓ per-call tier | ✗ | ✗ | ✗ | ✗ |
| Sparkline (last-10 tier mix) | ✓ inline | ✗ | ✗ | ✗ | ✗ |
| End-of-session digest | ✓ mooter digest | ✗ | ✗ | ✗ | ✗ |
| Live local subagent visibility | ✓ 🐄×N live | ✗ | ✗ | ✗ | ✗ |
| Cost tracking per-prompt | Not measured (no token logging) | ✗ | ✓ in logs | ⚠️ session | ✓ dashboard |
| Open source | ✓ MIT | n/a (closed) | ✓ Apache 2 | ✓ Apache 2 | ✗ (gateway hosted) |
| Free | ✓ (MIT license) | depends on sub | self-host or paid | ✓ | ⚠️ fee on credit purchases, no inference markup (openrouter.ai/docs/faq, read 2026-09-29) |
| Setup time | 1 command | n/a | 30+ min (config) | install plugin | sign up + key |
| Works without internet | ✓ T0 local | ✗ | ✗ | ✗ | ✗ |
Last updated unknown. Snapshot of public functionality at the time. We checked the docs. If we got something wrong, open an issue and we'll fix it.
Cursor / Copilot / Cody — and the open-source agents Cline, Aider and Roo Code — are AI coding assistants/agents, not LLM routers. They decide what to do; mooter decides which model runs each step. It complements them (run mooter under Claude Code while you use any of them elsewhere) rather than replacing them — so this table compares routers, not agents.
Snapshot reflects mooter v1.53.0. Capabilities and scores are derived, sourced, and open to correction.